Cohere: Command A+

Command A+ is a text generation model from Cohere, positioned by the provider as its flagship offering for enterprise agentic workflows. It belongs to Cohere's Command family and is aimed at teams building applications that need a capable general purpose language model behind an API. If you already use other Cohere models or you are evaluating providers for an agent style workload, this is the top tier option in that line.

On the input side, the model accepts both text and images, so prompts can combine written instructions with visual content. Output is text only. The context window is 192,000 tokens, which is large enough to hold long documents, extended conversation histories or sizeable tool outputs in a single request. The maximum output length is 64,000 tokens, so the model can produce long responses, not just short completions. It supports native tool calling with strict tool schemas and structured outputs, which means responses can be constrained to a schema you define rather than free form prose.

Those capabilities point to a few practical uses. Agentic workflows are the stated focus, so tool calling with strict schemas fits tasks where the model has to select functions, pass arguments and chain steps. Structured outputs suit data extraction from documents or images, form filling and any pipeline where a downstream system expects a fixed JSON shape. The combination of image input and a 192K context window works for document question answering, where a page image or long text is supplied alongside a question. Reasoning support through the reasoning and include_reasoning parameters also fits multi step problems where you want the model to work through intermediate steps.

You call it through api14's OpenAI compatible API using the model id "cohere/command-a-plus". Because the interface follows the OpenAI chat completions shape, existing OpenAI SDKs and client libraries generally work with only a base URL and key change. The request parameters supported are frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, response_format, seed, stop, structured_outputs, temperature, tools, top_k and top_p. That covers sampling controls, deterministic output via seed, stop sequences, output length via max_tokens, schema enforcement through response_format and structured_outputs, and tool definitions through tools.

There is no stated limitation in the facts beyond the text only output modality, so if you need image or audio generation this model is not the fit. For straightforward short text tasks, a smaller model may be sufficient, since the large context window and reasoning features matter most on long or complex inputs.